I tried it but I too felt it was clunky for the above reasons (3 being the fatal one). In addition:

• can't use it on a cluster that autoterminates/actually had to have a dedicated cluster for each user I believe
• the login process for the open source version is bad (allowing one user and having to share the credentials)
• persisting the workspace at all is a pain

All of this work is to avoid using a notebook that is tightly integrated with databricks. The reward isn't worth the effort.